NR PDCSH scrambling optimization (factor 10 reduction)

openair1/PHY/NR_TRANSPORT/nr_dlsch.c

@@ -65,6 +65,47 @@ void nr_pdsch_codeword_scrambling(uint8_t *in,
 
 }
 
+void nr_pdsch_codeword_scrambling_optim(uint8_t *in,
+					uint32_t size,
+					uint8_t q,
+					uint32_t Nid,
+					uint32_t n_RNTI,
+					uint32_t* out) {
+  
+  uint32_t x1, x2, s=0,in32;
+
+  x2 = (n_RNTI<<15) + (q<<14) + Nid;
+
+  s=lte_gold_generic(&x1, &x2, 1);
+
+#if defined(__AVX2__)
+  for (int i=0; i<((size>>5)+((size&0x1f) > 0 ? 1 : 0)); i++) {
+    in32=_mm256_movemask_epi8(_mm256_slli_epi16(((__m256i*)in)[i],7));
+    out[i]=(in32^s);
+    //    printf("in[%d] %x => %x\n",i,in32,out[i]);
+    s=lte_gold_generic(&x1, &x2, 0);
+  }
+#elif defined(__SSE4__)
+  _m128i *in128;
+  for (int i=0; i<((size>>5)+((size&0x1f) > 0 ? 1 : 0)); i++) {
+    in128=&((__m128i*)in)[i<<1];
+    ((uint16_t*)&in32)[0] = _mm128_movemask_epi8(_mm256_slli_epi16(in128[0],7));
+    ((uint16_t*)&in32)[1] = _mm128_movemask_epi8(_mm256_slli_epi16(in128[1],7));
+    out[i]=(in32^s);
+    s=lte_gold_generic(&x1, &x2, 0);
+  }
+  //#elsif defined(__arm__) || defined(__aarch64)
+  
+#else 
+  nr_pdsch_codeword_scrambling(in,
+			       size,
+			       q,
+			       Nid,
+			       n_RNTI,
+			       out);
+#endif
+}
+
 
 uint8_t nr_generate_pdsch(NR_gNB_DLSCH_t *dlsch,
                           uint32_t ***pdsch_dmrs,
@@ -133,7 +174,7 @@ uint8_t nr_generate_pdsch(NR_gNB_DLSCH_t *dlsch,
   for (int q=0; q<rel15->NrOfCodewords; q++)
     memset((void*)scrambled_output[q], 0, (encoded_length>>5)*sizeof(uint32_t));
   for (int q=0; q<rel15->NrOfCodewords; q++)
-    nr_pdsch_codeword_scrambling(harq->f,
+    nr_pdsch_codeword_scrambling_optim(harq->f,
 				       encoded_length,
 				       q,
 				       rel15->dlDmrsScramblingId,